LSTA: Long Short-Term Attention for Egocentric Action Recognition
This addresses the challenge of fine-grained discrimination in egocentric video analysis for applications like human-computer interaction, though it appears incremental as it builds on existing attention mechanisms.
The paper tackled the problem of egocentric activity recognition by proposing LSTA, a mechanism that focuses on spatial relevant parts and tracks attention smoothly across video sequences, achieving state-of-the-art performance on four standard benchmarks.
Egocentric activity recognition is one of the most challenging tasks in video analysis. It requires a fine-grained discrimination of small objects and their manipulation. While some methods base on strong supervision and attention mechanisms, they are either annotation consuming or do not take spatio-temporal patterns into account. In this paper we propose LSTA as a mechanism to focus on features from spatial relevant parts while attention is being tracked smoothly across the video sequence. We demonstrate the effectiveness of LSTA on egocentric activity recognition with an end-to-end trainable two-stream architecture, achieving state of the art performance on four standard benchmarks.